基于隐马尔可夫模型-径向基神经网络的表面肌电信号识别

A HMM-RBFN hybrid classifier for surface electromyography signals classification

目的:利用隐马尔可夫模型-径向基神经网络(HMM-RBFN)混合模型对7种手指动作进行辨识,探索控制HIT多自由度灵巧手的有效控制策略。方法:8例健康实验对象参加了试验,4例女性,4例男性。每例实验对象按提示完成7种手指动作,每种动作重复50次。通过表面肌电信号(sEMG)采集系统,提取实验对象前臂4块肌肉的sEMG,在对其进行预处理并提取小波变换特征向量后,分别送入HMM模型分类器及HMM-RBFN混合模型分类器进行训练。结果:HMM-RBFN混合模型识别效果和稳定性都大大优于HMM模型,验证了HMM-RBFN混合模型的有效性。结论:①HMM模型在sEMG识别中的效果没有其在语音信号识别中的好,有必要对其进行改进,以便更好的应用于sEMG的识别;②将HMM模型和神经网络组成混合分类器,可以弥补彼此的不足,获得更好的性能。

Objective：To classify surface electromyography（sEMG） signals by using HMM-RBFN hybrid classifier and to explore the strategy of effectively controlling hand prosthesis. Method：Eight subjects （male 4, female 4） with normal upper limbs were selected in the experiments. Each subject was instructed to perform 7 kinds of fingers movement and each motion was repeated 50 times. The sEMG signals were recorded on 4 forearm muscles. Features of sEMG signals were extracted using wavelet transform and conveyed to HMM classifier and HMM-RBFN hybrid classifier for training. Result： HMM-RBFN hybrid classifier provided better results than that from the single HMM classifier.Conclusion：①The performance of HMM classifier is not so excellent in sEMG signal discrimination. ②The HMM-RBFN hybrid classifier combine the advantages of two individual classifiers and offset their disadvantages,hence it achieves higher discrimination, accuracy and stability.